Every case is read from the corpus CSV on the server. Load any pair into the workspace and run the detector on it.